Core Concepts
Large Language Models (LLMs) can be enhanced for cognitive distortion classification through the ERD framework, improving performance and specificity.
Abstract
1. Abstract:
Using Large Language Models (LLMs) to improve psychotherapy accessibility.
Proposal of ERD framework for cognitive distortion classification.
2. Introduction:
LLMs' dominance in machine learning and AI.
Applications in medical domain and mental health support.
3. DoT Method Challenges:
Overdiagnosing cognitive distortions.
Limitations in multi-class setup performance.
4. ERD Framework Contributions:
Introduction of ERD framework with Extraction, Reasoning, and Debate steps using LLMs.
Performance improvements in F1 score and specificity.
5. Experiments:
Dataset details and experimental settings.
Performance enhancements with Extraction and Debate modules.
6. Conclusion:
Summary of ERD framework's effectiveness in cognitive distortion classification.
Stats
ERDは、歪曲分類タスクの多クラスF1スコアを9%以上向上させ、バイナリ特異性スコアを25%以上向上させる。